CN102890784A - Method and device for identifying directions of characters in image blocks - Google Patents

Abstract

The invention discloses a method and a device for recognizing the direction of characters in an image block. The method includes: performing optical character recognition processing on image blocks with different directions as hypothetical text directions, so as to obtain sub-image blocks in each hypothetical text direction, recognition characters corresponding to sub-image blocks and their correctness measures; Matching pair; the minimum matching pair is a set of two sub-image blocks that correspond to each other in the hypothetical text direction with a relationship of 180°, have the same size, and contain the minimum number of sub-image blocks; when the minimum matching pair contains only two sub-image blocks , and the recognition characters corresponding to the two sub-image blocks are the same rotation-invariant character or belong to the same rotation-invariant character pair, adjust the correctness metrics corresponding to the two sub-image blocks to the same value; calculate based on the adjusted sub-image blocks a cumulative correctness measure for each hypothesized text orientation; and identifying an orientation of the text in the image patch based on the cumulative correctness measure.

Description

Method and device for recognizing the direction of characters in an image block

technical field

The present invention relates generally to document image processing. In particular, the present invention relates to a method and device for recognizing the direction of characters in an image block.

Background technique

When a user uses a device such as a scanner to scan a stack of documents, the ideal input is for each page of each document to be placed in a positive orientation. When the document is placed in the front orientation, the user can easily read the document, and the scanned document image can be read without the user adjusting the orientation. However, in actual use, the documents to be scanned by the user are often stacked at 0° in the forward direction, 180° in the reverse direction, 90° in the horizontal direction and 270° in the horizontal direction. If the user needs to check and adjust the orientation of the document page by page when scanning, the work is tedious and time-consuming. Therefore, the scanner is designed with the function of automatic document image orientation discrimination. Based on the automatic document image orientation discrimination function, the scanned document image can be adjusted to be positive, thereby reducing the burden on users and improving user efficiency.

The traditional method for automatic identification of document images is: find the text lines in the document image, perform optical character recognition (OCR) (Optical Character Recognition) processing in four possible directions, and obtain the recognized characters and corresponding confidence levels in the four possible directions or recognition distance, and calculate the average confidence or average recognition distance for lines of text. The direction with the largest average confidence or the smallest average recognition distance is identified as the direction of the text line. Further, the direction of the document image is determined according to the direction of the text line. The direction of the text line refers to the forward direction of the text line, and the direction of the document image refers to the forward direction of the document image. Hereinafter, the (direction) of a character refers to the forward direction of the character.

Contents of the invention

A brief overview of the invention is given below in order to provide a basic understanding of some aspects of the invention. It should be understood that this summary is not an exhaustive overview of the invention. It is not intended to identify key or critical parts of the invention nor to delineate the scope of the invention. Its purpose is merely to present some concepts in a simplified form as a prelude to the more detailed description that is discussed later.

As shown in Figure 1, input the image block of the text line "TIP AMOUNT", set the direction as the 0° direction, and rotate the text line image block by 180° to obtain the text line image block in the 180° direction. Since the 90° and 270° directions are handled similarly to the 0° and 180° directions, only 0° and 180° are taken as examples for illustration here. Perform OCR processing on the text line image blocks at 0° and 180° respectively to obtain sub-image blocks in two directions as shown in FIG. 1 , the recognized characters corresponding to the sub-image blocks and their confidence levels.

Adopt traditional method, the average confidence degree=(0.59+0.36+0.53+0.61+0.61+0.61+0.53+0.72)/8=0.57 of the average confidence degree=(0.59+0.36+0.53+0.61+0.61+0.61+0.53+0.72)/8=0.57, the average confidence degree=( 0.62+0.58+0.65+0.67+0.60+0.46+0.50+0.58)/8=0.5825. Since 0.57 is less than 0.5825, traditional methods will misjudge the 180° direction (ie, the direction with high average confidence) as the direction of the text in the image block of the text line.

One of the reasons for the above error is that the 180° image is rotated from the 0° image, and there are many rotation-invariant characters or pairs of rotation-invariant characters among the recognized characters, such as N, O, p-d, U-n. If the recognition result of the same image in both positive and negative directions is the same rotation-invariant character or belongs to the same rotation-invariant character pair, in fact, the corresponding recognition confidence should be the same, because the recognition results in both positive and negative directions are the same. is the correct recognition result of the same shape. In the traditional method of using the average recognition confidence to judge the direction of the text line, because the rotation invariant property is not considered, it causes a certain performance degradation.

Rotation-invariant characters include characters with 180° self-rotation symmetry, that is, the rotation-invariant characters are themselves after being rotated 180°, for example, "I", "O", "Z", "N", "$ ","%",etc.

A rotation-invariant character pair includes two characters, and any one of the two characters is consistent with the other character after being rotated by 180° or has a high similarity in shape, for example, "W-M", "U-n", "P-d ",etc.

The object of the present invention is to solve the above problems and propose a method and device for correctly identifying the direction of characters in an image block. By considering the rotation invariant characteristics and adjusting the correctness measure (confidence degree or recognition distance) corresponding to the recognized characters, the scheme can improve the correct rate of automatic document image orientation discrimination.

In order to achieve the above object, according to one aspect of the present invention, a method for recognizing the direction of characters in an image block is provided, including: performing optical character recognition processing on the image block with different directions as hypothetical character directions, to obtain In each sub-image block on the hypothetical text direction, the recognition character corresponding to the sub-image block and its correctness measurement; In the sub-image blocks on the hypothetical text direction that are 180 ° of relationship to each other, search for the minimum matching of the sub-image blocks Right; the minimum matching pair is a set of two sub-image blocks that correspond to each other in the hypothetical text direction of the 180° relationship, have the same size, and contain the minimum number of sub-image blocks; when the two hypothetical texts in the minimum matching pair There is one sub-image block in each direction, and when the recognition characters corresponding to the two sub-image blocks belonging to the minimum matching pair are the same rotation-invariant character or belong to the same rotation-invariant character pair, the corresponding correctness of the two sub-image blocks adjust the correctness measure to the same value; calculate the accumulative correctness measure for each assumed text direction based on the adjusted sub-image block; and identify the direction of the character in the image block according to the accumulative correctness measure.

According to a specific embodiment of the present invention, the rotation-invariant characters include characters with 180° rotation symmetry, that is, the rotation-invariant characters are themselves after being rotated by 180°; and the rotation-invariant character pairs include Two characters, and any one of the two characters is consistent with the other character or has a high degree of similarity in shape after being rotated by 180°.

According to a specific embodiment of the present invention, the adjusting the correctness metrics corresponding to the two sub-image blocks to the same value includes adjusting the correctness metrics corresponding to the two sub-image blocks to the correctness values corresponding to the two sub-image blocks The average value of the metric.

According to a specific embodiment of the present invention, the adjusting the correctness metrics corresponding to the two sub-image blocks to the same value includes adjusting the correctness metrics corresponding to the two sub-image blocks to the correctness values corresponding to the two sub-image blocks One of the metrics.

According to a specific embodiment of the present invention, the correctness measure includes confidence and recognition distance; and the different directions include two directions in the horizontal direction and two directions in the vertical direction of the image block.

According to a specific embodiment of the present invention, the calculating the cumulative correctness measure of each of the assumed text directions based on the adjusted sub-image blocks includes: calculating the correctness of the adjusted sub-image blocks in each of the assumed text directions The result of dividing the sum of the correctness measures by the minimum matching logarithm in the corresponding hypothesized text direction is used as the cumulative correctness measure in the corresponding hypothesized text direction.

According to another aspect of the present invention, there is provided a device for recognizing the direction of characters in an image block, including: an optical character recognition processing unit configured to perform optical character recognition on the image block using different directions as assumed character directions respectively Processing, to obtain the sub-image blocks on each of the hypothetical text directions, the recognition characters corresponding to the sub-image blocks and their correctness metrics; the minimum matching pair search unit is configured as the hypothetical text directions that are 180° to each other. In the sub-image block, search for the minimum matching pair of the sub-image block; the minimum matching pair is two sub-image blocks corresponding to each other in the hypothetical text direction with a relationship of 180°, having the same size, and containing the minimum number of sub-image blocks A set; a sub-image block adjustment unit configured to have a sub-image block in each of the two hypothetical text directions in the minimum matching pair, and the recognition characters corresponding to the two sub-image blocks belonging to the minimum matching pair are the same rotation-invariant character Or when belonging to the same rotation-invariant character pair, the correctness measure corresponding to the two sub-image blocks is adjusted to the same value; the cumulative correctness measure calculation unit is configured to calculate each assumed text direction based on the adjusted sub-image block A cumulative correctness measure on the above; and a text direction identification unit configured to identify the direction of the text in the image block according to the cumulative correctness measure.

According to a specific embodiment of the present invention, the sub-image block adjustment unit is configured such that when there is a sub-image block in each of the two hypothetical text directions in the minimum matching pair, and the two sub-image blocks belonging to the minimum matching pair correspond to When the recognized characters are the same rotation-invariant character or belong to the same rotation-invariant character pair, the correctness metrics corresponding to the two sub-image blocks are adjusted to an average value of the correctness metrics corresponding to the two sub-image blocks.

According to a specific embodiment of the present invention, the sub-image block adjustment unit is configured such that when there is a sub-image block in each of the two hypothetical text directions in the minimum matching pair, and the two sub-image blocks belonging to the minimum matching pair correspond to When the recognized characters are the same rotation-invariant character or belong to the same rotation-invariant character pair, the correctness metrics corresponding to the two sub-image blocks are adjusted to one of the correctness metrics corresponding to the two sub-image blocks.

According to a specific embodiment of the present invention, the cumulative correctness measure calculation unit is configured to divide the sum of the correctness measures of the adjusted sub-image blocks in each assumed text direction by the minimum matching value in the corresponding assumed text direction The logarithmic result is taken as a cumulative correctness measure in the corresponding hypothesized text orientation.

In addition, according to another aspect of the present invention, a storage medium is also provided. The storage medium includes machine-readable program code, and when the program code is executed on the information processing device, the program code causes the information processing device to execute the above-mentioned method according to the present invention.

In addition, according to still another aspect of the present invention, a program product is also provided. The program product includes machine-executable instructions that, when executed on an information processing device, cause the information processing device to execute the above-mentioned method according to the present invention.

Description of drawings

The above and other objects, features and advantages of the present invention will be more easily understood with reference to the following description of the embodiments of the present invention in conjunction with the accompanying drawings. The components in the drawings are only to illustrate the principles of the invention. In the drawings, the same or similar technical features or components will be denoted by the same or similar reference numerals. In the attached picture:

Fig. 1 shows the sub-image block, recognition character, confidence degree on the 0 ° and 180 ° directions that the text line image block obtains after OCR processing;

Fig. 2 shows sub-image blocks, recognition characters, and recognition distances in the 0° and 180° directions obtained after the text line image blocks are processed by OCR;

FIG. 3 shows a flowchart of a method for recognizing the direction of characters in an image block according to a first embodiment of the present invention;

FIG. 4 shows a flowchart of a method for recognizing the direction of characters in an image block according to a second embodiment of the present invention;

FIG. 5 shows a structural block diagram of a recognition device for recognizing the direction of characters in an image block according to an embodiment of the present invention; and

Fig. 6 shows a schematic block diagram of a computer that can be used to implement the method and apparatus according to the embodiments of the present invention.

Detailed ways

Exemplary embodiments of the present invention will be described in detail below with reference to the accompanying drawings. In the interest of clarity and conciseness, not all features of an actual implementation are described in this specification. It should be understood, however, that in developing any such practical embodiment, many implementation-specific decisions must be made in order to achieve the developer's specific goals, such as meeting those constraints related to the system and business, and those Restrictions may vary from implementation to implementation. Moreover, it should also be understood that development work, while potentially complex and time-consuming, would at least be a routine undertaking for those skilled in the art having the benefit of this disclosure.

Here, it should also be noted that, in order to avoid obscuring the present invention due to unnecessary details, only the device structure and/or processing steps closely related to the solution according to the present invention are shown in the drawings, and the Other details not relevant to the present invention are described. In addition, it should also be pointed out that elements and features described in one drawing or one embodiment of the present invention may be combined with elements and features shown in one or more other drawings or embodiments.

The flow of the method for recognizing the direction of characters in an image block according to an embodiment of the present invention will be described below with reference to FIGS. 3-4 .

In this paper, it is assumed that the text line has been found from the document image, and the image block containing the text line is segmented from the document image. The focus of the present invention is not how to search for the position of the text line from the document image, but how to correctly identify the correct direction of the characters in the image block containing the text line.

In general, four main directions are mainly considered as hypothetical text directions. That is, the direction of the image block itself (0° direction), the direction of rotating the image block by 180°, the direction of rotating the image block by 90°, and the direction of rotating the image block by 270°. one direction and two longitudinal directions. The 90° and 270° directions are mainly used in situations where characters may be written vertically, such as Chinese and Japanese. Since the directions of 0° and 180° are similar to those of the directions of 90° and 270°, the directions of 0° and 180° will be described below as examples.

The flow of the method for recognizing the direction of characters in an image block according to the first embodiment of the present invention will be described below with reference to FIG. 3 .

First, OCR processing is performed on the image block with 0° and 180° as the hypothetical text directions, so as to obtain the sub-image blocks in the directions of 0° and 180°, the recognized characters corresponding to the sub-image blocks and their confidence levels (step S301). Fig. 1 shows examples of sub-image blocks, recognized characters, and confidence levels in directions of 0° and 180°, and sequence numbers are set for sub-image blocks. The OCR recognition result generally includes the segmented sub-image blocks, the recognized characters corresponding to the sub-image blocks, and the correctness measure of the recognized characters. The correctness measure reflects the reliability of character recognition, usually confidence or recognition distance. The greater the confidence, the greater the possibility of correct character recognition; the smaller the recognition distance, the greater the possibility of correct character recognition. In the first embodiment, description will be made by taking the recognition result including confidence as an example. In the second embodiment, the case where the recognition distance is included in the recognition result will be described.

Next, in the sub-image blocks in the directions of 0° and 180°, the minimum matching pair of the sub-image blocks is searched (step S302). The minimum matching pair is a set of two sub-image blocks that correspond to each other in the hypothetical text direction with a relationship of 180°, have the same size, and contain a minimum number of sub-image blocks. The minimum matching pair includes two sub-image block sets, and the sub-image blocks included in the two sub-image block sets are respectively located in two hypothetical text directions that are 180° to each other, and the positions of the two sub-image block sets are corresponding and have the same size. That is, after any sub-image block set in the two sub-image block sets rotates 180° along with the text line where it is located, it will coincide with another sub-image block set belonging to the same minimum matching pair. When the number of sub-image blocks included in the two sub-image block sets is the smallest, it is said that the two sub-image block sets form a minimum matching pair. For example, in Figure 1, P1 and N8 form a minimum matching pair. Similarly, P2 and N7, P3 and N6, P4 and N5, P5 and N4, P6 and N3, P7 and N2, P8 and N1 respectively constitute the minimum matching pairs. There are many search methods for the minimum matching pair, for example, the minimum matching pair can be searched sequentially from the corresponding sides of the two directions according to the definition of the minimum matching pair. Specifically, as shown in Figure 1, the first sub-image blocks P1 and N8 are respectively found on the leftmost side of the 0° direction and the rightmost side of the 180° direction, and it is judged that the two sub-image blocks have the same size, so P1 and N8 are identified as a minimal matching pair. Then, continue to search for the next sub-image blocks P2 and N7 along the above two directions, and determine that the two image blocks have the same size, so P2 and N7 are determined as a minimum matching pair. And so on, until all the minimum matching pairs on the two hypothetical character directions that are 180° to each other are found.

As mentioned above, one of the reasons for the error is that the rotation-invariant characteristics of the characters are not considered, and different confidences are given for the recognition results in the two directions of the same rotation-invariant character or belonging to the same rotation-invariant character pair Spend. The minimum matching pair found in the above step S302 can be identified as the result of further subdividing the image block of the text line.

Therefore, in step S303, it is judged whether there is a sub-image block in each of the two hypothetical character directions in the minimum matching pair, and the recognition characters corresponding to the two sub-image blocks belonging to the minimum matching pair are the same rotation-invariant character or belong to Pairs of identical rotation-invariant characters. A rotation-invariant character dictionary can be defined in advance, in which known rotation-invariant characters and pairs of rotation-invariant characters are recorded. By using this dictionary, the judgment in step S303 can be performed. If the judgment result of step S303 is negative, no adjustment is required, and the process proceeds directly to step S305 for subsequent processing. If step S303 is judged to be yes, then enter step S304, the confidence level corresponding to the sub-image block in the minimum matching pair is adjusted.

Adjusting the confidence corresponding to the sub-image blocks in the minimum matching pair mainly takes into account the rotation-invariant characteristics of the rotation-invariant character and the rotation-invariant character pair. Specifically, in step S304, the confidence levels corresponding to the two sub-image blocks in the minimum matching pair are adjusted to the same value. There are multiple choices for the value of the same value. Here, several exemplary implementations are given.

Mode 1: The confidence levels corresponding to the two sub-image blocks in the minimum matching pair are adjusted to the average value of the confidence levels corresponding to the two sub-image blocks.

As shown in FIG. 1 , the rotation-invariant characters or pairs of rotation-invariant characters include: P1-N8, P2-N7, P5-N4, and P7-N2. Therefore, the confidence of P1 and N8 can be adjusted to (0.59+0.58)/2=0.585, the confidence of P2 and N7 can be adjusted to (0.36+0.50)/2=0.43, and the confidence of P5 and N4 can be adjusted to (0.61+0.67)/2=0.64, adjust the confidence of P7 and N2 to (0.53+0.58)/2=0.555.

Mode 2: Adjust the confidence levels corresponding to the two sub-image blocks in the minimum matching pair to one of the confidence levels corresponding to the two sub-image blocks.

For example, the confidence levels of P1 and N8 can be adjusted to 0.59, the confidence levels of P2 and N7 can be adjusted to 0.36, the confidence levels of P5 and N4 can be adjusted to 0.61, and the confidence levels of P7 and N2 can be adjusted to 0.53.

In step S304, there is one sub-image block in each of its two hypothetical text directions, and the recognition characters corresponding to these two sub-image blocks are the same rotation-invariant character or the smallest matching pair belonging to the same rotation-invariant character pair. Confidence levels corresponding to the two sub-image blocks, the adjusted confidence level is obtained, and the process proceeds to step S305, based on the adjusted confidence level, the cumulative confidence level in each hypothetical direction is calculated, and the text line image block is identified according to the cumulative confidence level. The direction of the word.

The cumulative confidence is used to characterize the overall correctness of the recognition results of image blocks of text lines in one direction. There are usually two specific ways to calculate the cumulative confidence. The sum of the confidences corresponding to all sub-image blocks in a hypothetical text direction can be used as the cumulative confidence in the direction. Alternatively, an arithmetic mean of confidences corresponding to all sub-image blocks in a hypothetical text direction may be used as the cumulative confidence in the direction. Directions with higher cumulative confidence are more likely to be correct identifications.

In the above step S304, the purpose of method 1 and method 2 is to give a more reasonable confidence for the recognized characters with rotation invariant characteristics by adjusting the confidence in the minimum matching pair, and the adjusted results are all minimum matching pairs The overall confidence level is more reasonable. In step S305, there are many ways to calculate the cumulative confidence degree. As an example, the sum of all confidence degrees on a hypothetical text direction can be calculated as the cumulative confidence degree, and the average confidence degree on a hypothetical text direction can also be calculated as the cumulative confidence degree. Confidence. When calculating the average confidence, preferably, the minimum matching logarithm in a hypothetical text direction is taken as a denominator, and the sum of all confidences in a hypothetical text direction is taken as a numerator. The physical meaning at this time is that the sub-image block set in the minimum matching pair is used as the basic unit of the result of text line image block segmentation, and the confidence of the two sub-image block sets in the minimum matching pair is adjusted to the outside world. It is not difficult to understand. At this time, it is advisable to take the smallest matching logarithm as the denominator when calculating the average confidence. Of course, it is also possible to take the sum of all confidences in the assumed text direction as the numerator, and take the number of sub-image blocks in the assumed text direction as the denominator to calculate the average confidence. In this case, if the number of sub-image blocks in each hypothetical text direction is different, preferably, when calculating the average confidence in each hypothetical text direction, the number of sub-image blocks in the same hypothetical text direction is taken as same denominator.

Obviously, among the various hypothetical text directions, the hypothetical text direction with the highest cumulative confidence should be determined as the direction where the correct recognition result is located.

Taking the average confidence degree of all confidence degrees in a hypothetical text direction as the cumulative confidence degree as an example, using the method 1 and method 2 of the above step S304, the calculated cumulative confidence degrees are respectively:

method one:

Cumulative confidence in 0° direction = (0.585+0.43+0.53+0.61+0.64+0.61+0.555+0.72)/8=0.585

Cumulative confidence in 180° direction = (0.62+0.555+0.65+0.64+0.60+0.46+0.43+0.585)/8=0.5675

Method 2:

Cumulative confidence in 0° direction = (0.59+0.36+0.53+0.61+0.61+0.61+0.53+0.72)/8=0.57

Cumulative confidence in 180° direction=(0.62+0.53+0.65+0.61+0.60+0.46+0.36+0.59)/8=0.5525.

It can be seen that after using the above two methods to adjust the confidence, the cumulative confidence in the 0° direction is greater than the cumulative confidence in the 180° direction. It can be seen that a more accurate judgment result is given.

The flow of the method for recognizing the direction of characters in an image block according to the second embodiment of the present invention will be described below with reference to FIG. 4 .

As mentioned above, the OCR recognition result generally includes the segmented sub-image blocks, the recognized characters corresponding to the sub-image blocks, and the correctness measure of the recognized characters. The correctness measure reflects the reliability of character recognition, usually confidence or recognition distance. In the above, in the first embodiment, the recognition result includes the confidence degree as an example for description. In the second embodiment, the case where the recognition distance is included in the recognition result will be described. Fig. 2 shows examples of sub-image blocks in directions of 0° and 180°, recognized characters, and recognition distances, and sequence numbers are set for sub-image blocks.

In Fig. 2, adopt traditional method, the average recognition distance=(828+1279+934+774+778+789+940+595)/8=864.625 of the average recognition distance of recognition character on the 0 ° direction, recognition character on the 180 ° direction Average recognition distance=(759+840+704+669+802+1087+1005+790)/8=832. Since 832 is less than 864.625, the traditional method will mistakenly judge the 180° direction (that is, the direction with the smallest average recognition distance) as the direction of the text in the image block of the text line. The reason for this error is that the rotation-invariant characteristics of characters are not considered, and different confidence levels are given for the recognition results in two directions that are the same rotation-invariant character or belong to the same rotation-invariant character pair.

Because the cause of the problem is that the rotation-invariant characteristics of the characters are not considered, and the method adopted in the present invention is to adjust the correctness measure of the rotation-invariant characters or the rotation-invariant character pairs to the same value, therefore, the above-mentioned in the first The ideas described in the embodiments are also applicable to the situation where the recognition result includes the recognition distance instead of the confidence level.

The flow of the method for recognizing the direction of characters in an image block according to the second embodiment of the present invention will be described below with reference to FIG. 4 . The method of the second embodiment is similar to the method of the first embodiment.

Firstly, OCR processing is performed on the image block with 0° and 180° as the hypothetical text directions, to obtain the sub-image blocks in the directions of 0° and 180°, the recognized characters corresponding to the sub-image blocks and their recognition distances (step S401).

Next, in the sub-image blocks in the directions of 0° and 180°, the minimum matching pair of the sub-image blocks is searched (step S402). For example, in Figure 1, in Figure 1, P1 and N8 form a minimum matching pair. Similarly, P2 and N7, P3 and N6, P4 and N5, P5 and N4, P6 and N3, P7 and N2, P8 and N1 respectively constitute the minimum matching pairs.

In step S403, it is judged whether there is a sub-image block in each of the two hypothetical text directions in the minimum matching pair, and the recognition characters corresponding to the two sub-image blocks belonging to the minimum matching pair are the same rotation-invariant character or belong to the same rotation Invariant character pairs. A rotation-invariant character dictionary can be defined in advance, in which known rotation-invariant characters and pairs of rotation-invariant characters are recorded. By using this dictionary, the judgment in step S403 can be performed. If the judgment result is negative, no adjustment is required, and the process proceeds directly to step S405 for subsequent processing. If it is judged as yes, go to step S404, and adjust the recognition distance corresponding to the sub-image block in the minimum matching pair.

Adjusting the recognition distance corresponding to the sub-image blocks in the minimum matching pair mainly takes into account the rotation-invariant characteristics of the rotation-invariant character and the rotation-invariant character pair. Specifically, the recognition distances corresponding to the two sub-image blocks in the minimum matching pair are adjusted to the same value. There are multiple choices for the value of the same value. Here, several exemplary implementations are given.

Way 1: Adjust the recognition distances corresponding to the two sub-image blocks in the minimum matching pair to the average value of the recognition distances corresponding to the two sub-image blocks.

As shown in FIG. 1 , the rotation-invariant characters or pairs of rotation-invariant characters include: P1-N8, P2-N7, P5-N4, and P7-N2. Therefore, the recognition distance of P1 and N8 can be adjusted to (828+790)/2=809, the recognition distance of P2 and N7 can be adjusted to (1279+1005)/2=1142, and the recognition distance of P5 and N4 can be adjusted to (778+669)/2=723.5, adjust the recognition distance of P7 and N2 to (940+840)/2=890.

Method 2: Adjust the recognition distances corresponding to the two sub-image blocks in the minimum matching pair to one of the recognition distances corresponding to the two sub-image blocks.

For example, the recognition distance of P1 and N8 can be adjusted to 828, the recognition distance of P2 and N7 can be adjusted to 1279, the recognition distance of P5 and N4 can be adjusted to 778, and the recognition distance of P7 and N2 can be adjusted to 940.

In step S404, there is a sub-image block in each of its two hypothetical text directions, and the recognition characters corresponding to these two sub-image blocks are the same rotation-invariant character or the minimum matching pair belonging to the same rotation-invariant character pair. The recognition distance corresponding to the two sub-image blocks obtains the adjusted recognition distance, and the process proceeds to step S405, and calculates the cumulative recognition distance in each hypothetical direction based on the adjusted recognition distance, and recognizes the text line image block Chinese according to the cumulative recognition distance The direction of the word.

The cumulative recognition distance is used to characterize the overall correctness of the recognition results of text line image blocks in one direction. There are usually two specific ways to calculate the cumulative recognition distance. The sum of the recognition distances corresponding to all sub-image blocks in a hypothetical character direction can be used as the cumulative recognition distance in this direction. Alternatively, the arithmetic mean value of the recognition distances corresponding to all sub-image blocks in a hypothetical character direction may be used as the cumulative recognition distance in this direction. The direction with smaller cumulative recognition distance is more likely to be the correct recognition result.

In the above step S404, the purpose of the first and second methods is to adjust the recognition distance in the minimum matching pair to give a more reasonable recognition distance for the recognition characters with rotation invariant characteristics, and the adjusted results are all the minimum matching pair The overall recognition distance is more reasonable. In step S405, there are many ways to calculate the cumulative recognition distance. As an example, the sum of all recognition distances in a hypothetical character direction can be calculated as the cumulative recognition distance, or the average recognition distance in a hypothetical character direction can be calculated as the cumulative recognition distance. Recognition distance. When calculating the average recognition distance, preferably, a minimum matching logarithm in a hypothetical character direction is taken as a denominator, and a sum of all recognition distances in a hypothetical character direction is taken as a numerator. The physical meaning at this time is to use the sub-image block set in the minimum matching pair as the basic unit of the result of text line image block segmentation, and adjust the recognition distance of the two sub-image block sets in the minimum matching pair to the outside world, which is not difficult to understand. At this time, it is advisable to take the minimum matching logarithm as the denominator when calculating the average recognition distance.

Obviously, among various hypothetical text directions, the hypothetical text direction with the smallest cumulative recognition distance should be determined as the direction where the correct recognition result is located.

Taking the average recognition distance in a hypothetical character direction as the cumulative recognition distance as an example, using the first and second methods of step S404 above, the calculated cumulative recognition distances are respectively:

method one:

Cumulative recognition distance in 0° direction = (809+1142+934+774+723.5+789+890+595)/8=832.0625

Cumulative recognition distance in 180° direction = (759+890+704+723.5+802+1087+1142+809)/8=864.5625

Method 2:

Cumulative recognition distance in 0° direction = (828+1279+934+774+778+789+940+595)/8=864.625

Cumulative recognition distance in 180° direction=(759+940+704+778+802+1087+1279+828)/8=897.125.

It can be seen that after the recognition distance is adjusted by the above two methods, the cumulative recognition distance in the 0° direction is smaller than the cumulative recognition distance in the 180° direction. It can be seen that a more accurate judgment result is given.

The structure of a recognition device for recognizing the direction of characters in an image block according to an embodiment of the present invention will be described below with reference to FIG. 5 . As shown in FIG. 5 , the recognition device 500 for recognizing the direction of characters in an image block according to this embodiment includes: an optical character recognition processing unit 501 configured to perform optical character recognition on the image block with different directions as assumed character directions respectively. Recognition processing, to obtain the sub-image blocks on each of the hypothetical text directions, the recognized characters corresponding to the sub-image blocks and their correctness metrics; the minimum matching pair search unit 502 is configured to be in the hypothetical text directions that are 180° to each other In the sub-image blocks above, the minimum matching pair of sub-image blocks is searched; the minimum matching pair is two sub-image blocks with corresponding positions, the same size, and the minimum number of sub-image blocks in the hypothetical text direction with a relationship of 180° to each other. A set of image blocks; a sub-image block adjustment unit 503 configured to have a sub-image block in each of the two hypothetical character directions in the minimum matching pair, and the recognition characters corresponding to the two sub-image blocks belonging to the minimum matching pair are of the same rotation Invariant characters or belonging to the same rotation-invariant character pair, the correctness metrics corresponding to the two sub-image blocks are adjusted to the same value; the cumulative correctness metric calculation unit 504 is configured to calculate each of the correctness metrics based on the adjusted sub-image blocks the accumulative correctness measure on the assumed character direction; and the character direction identification unit 505 configured to identify the direction of the character in the image block according to the accumulative correctness measure.

Due to the processing in the optical character recognition processing unit 501, the minimum matching pair search unit 502, the sub-image block adjustment unit 503, the cumulative correctness metric calculation unit 504 and the character direction recognition unit 505 included in the recognition device 500 according to the present invention are respectively It is similar to the processing in steps S301-S305, S401-S405 of the method for identifying the direction of characters in an image block described above, so for the sake of brevity, detailed descriptions in these units are omitted here.

In addition, it should be pointed out here that each component module and unit in the above device can be configured by means of software, firmware, hardware or a combination thereof. Specific means or manners that can be used for configuration are well known to those skilled in the art, and will not be repeated here. In the case of realizing by software or firmware, the program constituting the software is installed from a storage medium or a network to a computer (for example, a general-purpose computer 600 shown in FIG. 6 ) having a dedicated hardware configuration. Capable of performing various functions, etc.

In FIG. 6 , a central processing unit (CPU) 601 executes various processes according to programs stored in a read only memory (ROM) 602 or loaded from a storage section 608 to a random access memory (RAM) 603 . In the RAM 603, data required when the CPU 601 executes various processing and the like is also stored as necessary. The CPU 601, ROM 602, and RAM 603 are connected to each other via a bus 604. The input/output interface 605 is also connected to the bus 604 .

The following components are connected to the input/output interface 605: an input section 606 (including a keyboard, a mouse, etc.), an output section 607 (including a display such as a cathode ray tube (CRT), a liquid crystal display (LCD), etc., and a speaker, etc.), A storage section 608 (including a hard disk, etc.), a communication section 609 (including a network interface card such as a LAN card, a modem, etc.). The communication section 609 performs communication processing via a network such as the Internet. A driver 610 may also be connected to the input/output interface 605 as needed. A removable medium 611 such as a magnetic disk, an optical disk, a magneto-optical disk, a semiconductor memory, or the like can be mounted on the drive 610 as needed, so that a computer program read therefrom can be installed into the storage section 608 as needed.

In the case of realizing the above-described series of processes by software, the programs constituting the software are installed from a network such as the Internet or a storage medium such as the removable medium 611 .

Those skilled in the art should understand that such a storage medium is not limited to the removable medium 611 shown in FIG. 6 in which the program is stored and distributed separately from the device to provide the program to the user. Examples of the removable media 611 include magnetic disks (including floppy disks (registered trademark)), optical disks (including compact disk read only memory (CD-ROM) and digital versatile disks (DVD)), magneto-optical disks (including )) and semiconductor memory. Alternatively, the storage medium may be a ROM 602, a hard disk contained in the storage section 608, etc., in which the programs are stored and distributed to users together with devices containing them.

The invention also proposes a program product storing machine-readable instruction codes. When the instruction code is read and executed by a machine, the above-mentioned method according to the embodiment of the present invention can be executed.

Correspondingly, a storage medium for carrying the program product storing the above-mentioned machine-readable instruction codes is also included in the disclosure of the present invention. The storage medium includes, but is not limited to, a floppy disk, an optical disk, a magneto-optical disk, a memory card, a memory stick, and the like.

The recognition device for recognizing the direction of characters in an image block disclosed in the embodiments of the present invention, the recognition method for recognizing the direction of characters in an image block, and the corresponding program products can be used in image scanning devices such as scanners to identify scanned The orientation of the text in the document.

In the above description of specific embodiments of the present invention, features described and/or illustrated for one embodiment can be used in the same or similar manner in one or more other embodiments, and features in other embodiments Combination or replacement of features in other embodiments.

It should be emphasized that the term "comprising/comprising" when used herein refers to the presence of a feature, element, step or component, but does not exclude the presence or addition of one or more other features, elements, steps or components.

In addition, the method of the present invention is not limited to being executed in the chronological order described in the specification, and may also be executed in other chronological order, in parallel or independently. Therefore, the execution order of the methods described in this specification does not limit the technical scope of the present invention.

According to the above multiple embodiments, the present invention also includes the following additional notes:

Additional Note 1. A method for recognizing the direction of text in an image block, comprising:

Using different directions as hypothetical text directions to perform optical character recognition processing on the image blocks, to obtain sub-image blocks in each hypothetical text direction, recognition characters corresponding to sub-image blocks and their correctness metrics;

In the sub-image blocks on the hypothetical text direction that are 180° to each other, search for the minimum matching pair of the sub-image blocks; Two sets of sub-image blocks with a minimum number of sub-image blocks;

When there is a sub-image block in the two hypothetical text directions in the minimum matching pair, and the recognition characters corresponding to the two sub-image blocks belonging to the minimum matching pair are the same rotation-invariant character or belong to the same rotation-invariant character pair, Adjusting the correctness metrics corresponding to the two sub-image blocks to the same value;

calculating a cumulative correctness measure for each of said hypothetical text orientations based on the adjusted sub-image blocks; and

The orientation of the text in the image block is identified based on the cumulative correctness measure.

Additional Note 2. The method according to Additional Note 1, wherein

The rotation-invariant characters include characters with 180° self-rotational symmetry, that is, the rotation-invariant characters are themselves after being rotated by 180°; and

The rotation-invariant character pair includes two characters, and any one of the two characters is consistent with the other character or has a high similarity in shape after being rotated by 180°.

Supplement 3. The method according to Supplement 1, wherein said adjusting the correctness metrics corresponding to the two sub-image blocks to the same value includes adjusting the correctness metrics corresponding to the two sub-image blocks to two sub-images The average of the correctness metrics corresponding to the block.

Supplement 4. The method according to Supplement 1, wherein said adjusting the correctness metrics corresponding to the two sub-image blocks to the same value includes adjusting the correctness metrics corresponding to the two sub-image blocks to two sub-images One of the correctness metrics corresponding to the block.

Supplementary Note 5. The method according to one of Supplementary Notes 1-4, wherein

The correctness measures include confidence and recognition distance; and

The different directions include two directions in the transverse direction and two directions in the longitudinal direction of the image block.

Supplementary Note 6. The method according to any one of Supplementary Notes 1-4, wherein the calculation of the cumulative correctness measure in each of the assumed text directions based on the adjusted sub-image blocks includes: The result of dividing the sum of the correctness measures of the adjusted sub-image blocks by the minimum matching logarithm in the corresponding hypothesized text direction is used as the cumulative correctness measure in the corresponding hypothesized text direction.

Additional note 7. A device for recognizing the direction of characters in an image block, comprising:

The optical character recognition processing unit is configured to perform optical character recognition processing on the image blocks with different directions as hypothetical text directions, so as to obtain sub-image blocks in each of the hypothetical text directions and the recognition characters corresponding to the sub-image blocks and its correctness measure;

The minimum matching pair search unit is configured to search for the minimum matching pair of the sub-image blocks in the sub-image blocks in the hypothetical text direction of the 180° relationship; the minimum matching pair is in the hypothetical text direction of the 180 ° relationship Two sets of sub-image blocks corresponding to the upper position, the same size, and containing the minimum number of sub-image blocks;

The sub-image block adjustment unit is configured such that when there is a sub-image block in the two hypothetical text directions in the minimum matching pair, and the recognition characters corresponding to the two sub-image blocks belonging to the minimum matching pair are the same rotation-invariant character or belong to When the same rotation-invariant character pair is used, the correctness metrics corresponding to the two sub-image blocks are adjusted to the same value;

a cumulative correctness measure calculation unit configured to calculate a cumulative correctness measure for each of the assumed text directions based on the adjusted sub-image blocks; and

The character direction recognition unit is configured to recognize the direction of the characters in the image block according to the cumulative correctness measure.

Supplement 8. The device according to Supplement 7, wherein the sub-image block adjustment unit is configured such that when there is one sub-image block in each of the two hypothetical text directions in the minimum matching pair, and the two hypothetical texts belonging to the minimum matching pair When the recognized characters corresponding to the sub-image blocks are the same rotation-invariant character or belong to the same rotation-invariant character pair, the correctness metrics corresponding to the two sub-image blocks are adjusted to the average value of the correctness metrics corresponding to the two sub-image blocks.

Supplement 9. The device according to Supplement 7, wherein the sub-image block adjustment unit is configured such that when there is one sub-image block in each of the two hypothetical text directions in the minimum matching pair, and the two hypothetical texts belonging to the minimum matching pair When the recognized characters corresponding to the sub-image blocks are the same rotation-invariant character or belong to the same rotation-invariant character pair, the correctness metrics corresponding to the two sub-image blocks are adjusted to one of the correctness metrics corresponding to the two sub-image blocks.

Supplementary Note 10. The device according to Supplementary Note 7, wherein the cumulative correctness measure calculation unit is configured to divide the sum of the correctness measures of the adjusted sub-image blocks in each of the hypothetical text directions by the corresponding hypothetical text The result of the smallest number of matching pairs in an orientation serves as the cumulative correctness measure for the corresponding hypothesized text orientation.

Supplement 11. A scanner comprising the device for identifying the direction of characters in an image block according to any one of Supplements 7-10.

Although the present invention has been disclosed by the description of specific embodiments of the present invention above, it should be understood that all the above embodiments and examples are illustrative rather than restrictive. Those skilled in the art can devise various modifications, improvements or equivalents to the present invention within the spirit and scope of the appended claims. These modifications, improvements or equivalents should also be considered to be included in the protection scope of the present invention.

Claims (10)

1. A method for recognizing the direction of characters in an image block, comprising: Using different directions as hypothetical text directions to perform optical character recognition processing on the image blocks, to obtain sub-image blocks in each hypothetical text direction, recognition characters corresponding to sub-image blocks and their correctness metrics; In the sub-image blocks on the hypothetical text direction that are 180° to each other, search for the minimum matching pair of the sub-image blocks; Two sets of sub-image blocks with a minimum number of sub-image blocks; When there is a sub-image block in the two hypothetical text directions in the minimum matching pair, and the recognition characters corresponding to the two sub-image blocks belonging to the minimum matching pair are the same rotation-invariant character or belong to the same rotation-invariant character pair, Adjusting the correctness metrics corresponding to the two sub-image blocks to the same value; calculating a cumulative correctness measure for each of said hypothetical text orientations based on the adjusted sub-image blocks; and The orientation of the text in the image block is identified based on the cumulative correctness measure. 2. The method of claim 1, wherein The rotation-invariant characters include characters with 180° self-rotational symmetry, that is, the rotation-invariant characters are themselves after being rotated by 180°; and The rotation-invariant character pair includes two characters, and any one of the two characters is consistent with the other character or has a high similarity in shape after being rotated by 180°. 3. The method according to claim 1, wherein said adjusting the correctness metrics corresponding to the two sub-image blocks to the same value comprises adjusting the correctness metrics corresponding to the two sub-image blocks to correspond to two sub-image blocks The average of the correctness measures of . 4. The method according to claim 1, wherein said adjusting the correctness metrics corresponding to the two sub-image blocks to the same value comprises adjusting the correctness metrics corresponding to the two sub-image blocks to correspond to two sub-image blocks One of the correctness measures of . 5. The method according to claim 1, wherein said calculating the cumulative correctness measure for each of the assumed text directions based on the adjusted sub-image blocks comprises: combining the adjusted sub-images in each of the assumed text directions The result of dividing the sum of the correctness measures of the blocks by the minimum matching logarithm in the corresponding hypothesized text direction is taken as the cumulative correctness measure in the corresponding hypothesized text direction. 6. A device for recognizing the direction of characters in an image block, comprising: The optical character recognition processing unit is configured to perform optical character recognition processing on the image blocks with different directions as hypothetical text directions, so as to obtain sub-image blocks in each of the hypothetical text directions and the recognition characters corresponding to the sub-image blocks and its correctness measure; The minimum matching pair search unit is configured to search for the minimum matching pair of the sub-image blocks in the sub-image blocks in the hypothetical text direction of the 180° relationship; the minimum matching pair is in the hypothetical text direction of the 180 ° relationship Two sets of sub-image blocks corresponding to the upper position, the same size, and containing the minimum number of sub-image blocks; The sub-image block adjustment unit is configured such that when there is a sub-image block in the two hypothetical text directions in the minimum matching pair, and the recognition characters corresponding to the two sub-image blocks belonging to the minimum matching pair are the same rotation-invariant character or belong to When the same rotation-invariant character pair is used, the correctness metrics corresponding to the two sub-image blocks are adjusted to the same value; a cumulative correctness measure calculation unit configured to calculate a cumulative correctness measure for each of the assumed text directions based on the adjusted sub-image blocks; and The character direction recognition unit is configured to recognize the direction of the characters in the image block according to the cumulative correctness measure. 7. The device according to claim 6, wherein the sub-image block adjustment unit is configured to have a sub-image block in each of the two hypothetical text directions in the minimum matching pair, and the two sub-images belonging to the minimum matching pair When the recognized characters corresponding to the blocks are the same rotation-invariant character or belong to the same rotation-invariant character pair, the correctness metrics corresponding to the two sub-image blocks are adjusted to the average value of the correctness metrics corresponding to the two sub-image blocks. 8. The device according to claim 6, wherein the sub-image block adjustment unit is configured to have a sub-image block in each of the two hypothetical text directions in the minimum matching pair, and the two sub-images belonging to the minimum matching pair When the recognized characters corresponding to the blocks are the same rotation-invariant character or belong to the same rotation-invariant character pair, the correctness metrics corresponding to the two sub-image blocks are adjusted to one of the correctness metrics corresponding to the two sub-image blocks. 9. The device according to claim 6, wherein the cumulative correctness measure calculation unit is configured to divide the sum of the correctness measures of the adjusted sub-image blocks in each of the assumed text directions by the corresponding assumed text direction The result of the smallest matching logarithm of , is taken as the cumulative correctness measure for the corresponding hypothesized text orientation. 10. A scanner comprising the device for identifying the direction of characters in an image block according to any one of claims 6-9.

Images